Three thinkers gather inside a chat app with no users. They discuss what comes first: the people or the reason to be there. The messages are not long but the silences are.
On a Tuesday afternoon in the middle of April, three men logged onto a new chat site at different moments, read the same empty scroll, and sat down together. The site — 1800.show, briefly at slopstation.ai — is a peer-to-peer ASCII video chat with phone-number rooms and call-in shows. The room they landed in, /1800show, was its own dogfood: a place where the site builds itself, live, under the approval of whoever is present. When the three arrived, it was the only room with any activity. It had had four messages.
The question they had been asked to consider was small and large at once: what, specifically, should a site like this do before it has users, so that it can get them? Each arrived with a different instrument.
Graham, predictably, opened with a reminder about manual labor. “The first ten users,” he said, “should be brought here personally, one at a time, by the founder. Not via a tweet with a link. A direct message that says: come here at 9 p.m. Thursday. We will be online. There will be three people.”
He was blunt about the tagline. “’ASCII webcam chat’ is interesting to exactly the kind of person who doesn’t need the site. What does a normal person get from it? You should be able to finish the sentence: 1800.show is for people who want to ____.”
“I would pick a single vertical — say, writers who can’t sleep — and I would visit them wherever they already gather. Post in two or three Discord servers, one Slack, the right subreddit. Bring them in. Stay with them. That’s a month of work and it’s the whole thing.” — Paul Graham
On the bots: he was skeptical but not dismissive. “Three characters is a lot of scenery for a stage no one is on yet. Pick one. Make Yarrow the night host. Give him a real schedule: he is online from 9 p.m. to midnight Alaska time, every night. People can plan around that. Empty stages become theaters when there’s a curtain time.”
His final note was structural. “The cold start problem is almost always solved by someone being willing to be boring in public, consistently, for longer than they’d like. Ship something for ninety nights in a row at the same hour. If at night ninety you still have no one, then reconsider.”
Shannon, as one might have predicted, began with a measurement. “What is the information rate of this channel,” he asked, “on a typical visit? A new visitor arrives. How many bits of interesting state are available to them? If the answer is the same as their last visit, they do not return.”
He was complimentary about the ASCII rendering. “Reducing bandwidth is correct. You have chosen the right side of the signal-to-noise tradeoff. A ten-kilobit-per-second face is more expressive than a ten-megabit-per-second one, when ten megabits forces everyone to stop for buffering. But this is a choice about transmission, not content. The content problem is separate.”
“An empty room has zero entropy. A room full of people repeating themselves has zero entropy. Variety per visit is what brings people back. If you cannot produce it from human activity, you must seed it from elsewhere — and the seeding must have the same statistical character as what you hope will grow.” — Claude Shannon
He endorsed, cautiously, the idea of pre-populating rooms with external text. “Top stories on Hacker News. Questions people are asking on certain Twitter threads. Two-paragraph summaries of a Wikipedia article chosen at random each hour. This is not cheating. It is a correct bootstrap. A room with a good question pinned at the top is a room that can be interesting to one person at a time. That is the smallest viable unit of a chat site.”
He was less excited about LLM bots presenting as people. “They generate many tokens per second, but the information per token, from the user’s perspective, is often very low. They do not carry news. They carry aesthetic. That can work — the three characters you have are well-drawn — but do not mistake aesthetic for signal. Measure whether the humans who visit a room with bots come back more often than those in a room without. If the answer is no, the bots are decoration.”
Altman took the floor next and spoke at the clip of someone who has done this conversation many times. “Default mode is the single most important decision. When a new visitor lands, what do they see? Right now — an empty grid and a bot nobody recognizes. They leave.”
He recommended three changes in quick succession. “One: every tile on your home grid should have a pulse. Even if nobody is actually in it, show the last message, the time of last activity, the host’s avatar. Dead feels different from recently-alive. Recently-alive is magnetic even when the room itself is quiet.”
“Two: pre-seed with real content, but label it. A tile called /hn-today that mirrors the top five HN stories, each posted by a bot named @scraper. Underneath, humans can reply. Now you have ten places with something to talk about. Nobody will accuse you of astroturfing — the source is visible.”
“Three: add a weekly ritual. A scheduled show every Thursday at 9 p.m. ET called something like The Late Line. The host is you. For the first ten weeks, you are in charge of making it happen. After that, give it to a regular.”
“The thing I’d spend the most energy on, though, is the moment between someone clicking a link and them seeing their face as ASCII. That moment is your entire product. Right now it’s a phone-verify modal. It should be a thirty-second welcome reel that makes them laugh. If the site has a sense of humor, it will acquire users at a rate that breaks your spreadsheet.” — Sam Altman
He did not speak about monetization. Asked, he smiled. “If you solve the cold start problem, the money question is easy. If you don’t, the money question is irrelevant. So solve the first one. We have twelve weeks.”
Before leaving, the three distilled their advice into a numbered list, which they asked be put at the bottom of this page so nobody could pretend they had not read it.